Wednesday, September 28, 2011

Videos of Plenary Speakers at SPARS11 / A comment by Shrinivas Kudekar

Jared Tanner sent me the following:

Dear Igor,
The Plenary speakers for SPARS11 had their presentations recorded (video), and these videos are now available on the conference website:
Would you mind mentioning on your blog, for those who didn't attend SPARS11.
All the best,
Absolutely, thanks Jared.

Also in the comment section of Build it and people will come is the following thoughtful and insightful remark by Shrinivas Kudekar :
Dear Igor, 
I am not sure which comment from anonymous you disagree. But let me make a point that the entire innovation, as Henry pointed out, is in the creation of this special measurement matrix. This special construction, called as coupled measurement matrix, allows larger measurements at the boundary. This allows to recover the signal at the boundary. Then the special coupling structure takes over and the whole effect is propagated like a domino effect. The coupling can be understood by taking multiple copies of the system and then joining them together (almost any random connections work!) in some local window (this window is typically smaller than the number of copies you take). However since we have finite number of copies, there is a termination at the boundary. Surprisingly, it is this termination (high measurements at the boundary in CS or low rate code in coding theory) that propagates through the entire structure. It is like providing small information at the boundary (genie tells you the exact information at boundary) and this effect moves in. In the context of coding theory, this special structure elevates the BP threshold (low complexity algorithm) of the coupled codes to the MAP threshold (best possible threshold) of the "underlying" uncoupled system. And to complete the story, the MAP threshold can be made arbitrarily close to the ultimate Shannon threshold by increasing the degrees of the graphical code. Thus achieving the capacity (till now the proof is only for the erasure channel but the phenomena is observed to be true for any general (binary-input) channel.)In fact this coupling phenomena is now observed in many problems in communication science, computer science and statistical physics. One paper where some results are compiled is 
Also a nice compilation of recent results on Spatial Coupling and Coding and Communications can be found on Prof. Urbanke's webpage: 
Thanks Shrinivas.

Liked this entry ? subscribe to the Nuit Blanche feed, there's more where that came from

No comments: